Institutional review board approval was obtained for this prospective
study. All patients who presented to a major tertiary-care children's hospital
between 1997 and 2002 with an acutely irritable hip and a differential
diagnosis of transient synovitis or septic arthritis were managed on the basis
of the previously published clinical practice
guideline30.
As in the previous
study27, the
diagnoses of septic arthritis and transient synovitis were operationally
defined on the basis of the white blood-cell count in the joint fluid, the
results of cultures of joint fluid and blood, and the clinical course. The
diagnosis of true septic arthritis was assigned when the patient had either a
positive finding on culture of joint fluid or a white blood-cell count in the
joint fluid of =50,000 cells/mm3 (=50.0 ×
109/L) with a positive finding on blood culture. The diagnosis of
presumed septic arthritis was explicitly assigned when the patient had a white
blood-cell count in the joint fluid of =50,000 cells/mm3
(=50.0 × 109/L) with negative findings on cultures of the
joint aspirate and blood. Thus, the group with septic arthritis included both
the group with true septic arthritis and the group with presumed septic
arthritis. The diagnosis of transient synovitis was explicitly assigned when
the patient had a white blood-cell count in the joint fluid of <50,000
cells/mm3 (<50.0 × 109/L) with negative
findings on culture, resolution of symptoms without antimicrobial therapy, and
no further development of a disease process as documented in the medical
record. The mean duration of follow-up was 11.8 months (range, 5.9 to 23.7
months).
Of the 213 eligible consecutive patients who were evaluated during the
study period, twenty-four were diagnosed with true septic arthritis,
twenty-seven were diagnosed with presumed septic arthritis, 103 were diagnosed
with transient synovitis, and fifty-nine were excluded. The exclusion criteria
were the same as those described in the previous
study27. The
fifty-nine excluded patients included individuals in atypical groups, such as
those with immunocompromise (ten patients), renal failure (three), neonatal
sepsis (three), postoperative infection of the hip (two), later development of
rheumatologic disease (one), or later development of Legg-Calvé-Perthes
disease (one). Two patients with septic arthritis and associated proximal
femoral osteomyelitis who had had symptoms for more than two weeks and who had
proximal femoral radiolucency on radiographs, an intraosseous abscess that was
confirmed at the time of arthrotomy and femoral neck drilling, and bacteremia
were excluded because they were not believed to pose the typical diagnostic
dilemma between septic arthritis and transient synovitis. To avoid information
bias associated with incomplete data analysis and to avoid selection bias
associated with the inclusion of patients with presumptive and inconsistent
diagnoses, a patient was excluded if joint fluid had not been obtained (or
insufficient fluid had been obtained) for a cell count, gram stain, or culture
(twenty-seven patients); if peripheral blood had not been obtained for a cell
count or culture (five); if the white blood-cell count in the joint fluid was
<50,000 cells/mm3 (<50.0 × 109/L) with
negative cultures but the patient was treated with arthrotomy and intravenous
administration of antibiotics (four); or if the white blood-cell count in the
joint fluid was <50,000 cells/mm3 (<50.0 ×
109/L) with negative cultures but the patient was managed with
intravenous administration of antibiotics alone on the pediatrics service
(one). Of the twenty-seven patients who were excluded because of the lack of
joint-fluid analysis, twenty-five had an insufficient quantity of aspirate
available for both a cell count and gram stain with culture, and therefore
only one test had been performed.
Data were obtained for all patients on age, gender, date of presentation,
duration of symptoms, history of fever, history of chills, weight-bearing
status, history of trauma, history of concurrent or recent infection, history
of recent antibiotic use, temperature, erythrocyte sedimentation rate, serum
white blood-cell count and differential, platelet count, hematocrit, results
of blood culture, evidence of hip joint effusion on radiographs, and results
of gram-staining, cell count, differential, and culture of joint fluid.
C-reactive protein values were not obtained for all patients because, during
the study period at our institution, the availability of C-reactive protein
testing evolved from weekly testing to once-daily testing to routine testing.
A history of fever was operationally defined as an oral temperature of
>38.5°C during the week before presentation. A history of chills was
operationally defined as positive documentation of chills; a negative history
of chills was coded either for documentation of no chills or for no
documentation of chills. Weight-bearing status was determined on the basis of
the clinical history and was considered the inability or refusal to bear
weight even with support. An effusion was defined as a side-to-side distance
of >2 mm from the medial part of the femoral head to the medial part of the
acetabulum on an anteroposterior pelvic radiograph.
Univariate analysis was performed with use of the two-sample Student t test
for continuous variables and with use of the Fisher exact test for categorical
variables. Comparisons were made between the group with true septic arthritis
and the group with presumed septic arthritis and between the group with septic
arthritis and the group with transient synovitis. Stepwise multiple logistic
regression with use of backward selection was performed to identify
independent clinical predictors, and comparisons were made between the septic
arthritis and transient synovitis groups. Variables associated with a p value
of <0.20 in the univariate analysis were chosen as candidates for the
multivariate model, with significance determined by the likelihood ratio
chi-square test. Regression model fit was estimated with the Hosmer-Lemeshow
goodness-of-fit test. Adjusted odds ratios and 95% confidence intervals were
derived with the method of maximum likelihood. A receiver operating
characteristic curve was constructed to assess the diagnostic performance of
the group of multivariate predictors in identifying septic arthritis.
Statistical analysis was performed with use of the SPSS (version 11.0; SPSS,
Chicago, Illinois) and SAS (version 6.12; SAS Institute, Cary, North Carolina)
software packages.
To evaluate the diagnostic performance of the previously described clinical
prediction in the current patient population, the multivariate predictors from
the current patient population were compared with those from the original
population. In addition, the predicted probability of septic arthritis from
the prediction rule was compared with actual distributions in the current
patient population. The area under the receiver operating characteristic curve
for the current patient population was compared with that for the original
population. The receiver operating characteristic curve is a graphic
analytical technique that is used to evaluate the diagnostic performance of a
test or a prediction rule. Sensitivity is plotted on the y axis, and the
false-positive rate (1 — specificity) is plotted on the x axis. The area
under the receiver operating characteristic curve is a summary measure of the
diagnostic performance of the prediction rule. A perfect rule would
approximate the upper left corner of the graph with an area under the curve of
1.0. Random guessing would be a straight line graph with an area under the
curve of 0.5.
Descriptive Data
Of the fifty-one patients with septic arthritis, twenty-four (47%) had
positive results on culture and twenty-seven (53%) had negative results on
culture. Of the twenty-four patients with positive results, sixteen had a
positive joint-fluid culture and a positive blood culture, six had a positive
joint-fluid culture and a negative blood culture, and two had a negative
joint-fluid culture and a positive blood culture. Organisms isolated on
culture included Staphylococcus aureus (fifteen patients),
Streptococcus pneumoniae (six), Neisseria meningitidis
(two), and group-A Streptococcus (one). Of the twenty-two patients with a
positive culture of joint fluid, seventeen had positive gram stains of joint
fluid. There were no positive gram stains of joint fluid from the patients who
had negative joint-fluid cultures.
Univariate Analysis: True Septic Arthritis Compared with Presumed
Septic Arthritis
The twenty-four patients who had true septic arthritis differed
significantly from the twenty-seven patients who had presumed septic arthritis
with regard to a history of recent antibiotic use (two patients [8%] compared
with ten patients [37%], p < 0.001), a history of chills (six patients
[25%] compared with zero patients [0%], p < 0.001), temperature elevation
(38.1°C ± 0.8°C compared with 37.4°C ± 0.9°C, p
= 0.02), erythrocyte sedimentation rate (52.9 ± 20.1 compared with 41.1
± 19.9 mm/hr, p = 0.04), a history of fever (twenty-three patients
[96%] compared with nineteen patients [70%], p = 0.005), male gender (fifteen
patients [63%] compared with seven patients [26%], p = 0.03), and serum white
blood-cell differential (bands [8.2% ± 9.4% compared with 3.8% ±
2.5%], p = 0.02). There were no significant differences (p > 0.05) with
regard to age, duration of symptoms, effusion on radiographs,
non-weight-bearing status, hematocrit, platelet count, serum white blood-cell
count, or serum white blood-cell differential for neutrophils, lymphocytes,
monocytes, atypical lymphocytes, eosinophils, and basophils.
Univariate Analysis: Septic Arthritis Compared with Transient
Synovitis
The patients who had septic arthritis differed significantly from those who
had transient synovitis with regard to gender, a history of fever,
non-weight-bearing status, temperature, erythrocyte sedimentation rate,
hematocrit, serum white blood-cell count, and lymphocytes (p < 0.05 for
all). There were no significant differences between these groups with regard
to age, duration of symptoms, chills, antibiotic use, radiographic evidence of
effusion, platelet count, neutrophils, monocytes, bands, atypical lymphocytes,
eosinophils, or basophils (p > 0.05) (see Appendix).
Multivariate Analysis: Septic Arthritis Compared with Transient
Synovitis
We identified the same four independent multivariate predictors of septic
arthritis in the current population as we had in the derivation population: a
history of fever, non-weight-bearing, an erythrocyte sedimentation rate of 40
mm/hr, and a serum white blood-cell count of >12,000 cells/mm3
(>12.0 × 109/L) (Table
I). The Hosmer-Lemeshow goodness-of-fit test revealed no
significant departure from good model fit (p > 0.05).
Algorithm for Probability of Septic Arthritis
The actual distribution of septic arthritis in the current population was
similar to the predicted probability of septic arthritis derived from the
original population for the algorithm based on all sixteen combinations of the
four predictors and for the simplified algorithm based on the number of
predictors (see Appendix).
For a patient with zero predictors, the predicted probability of septic
arthritis from the previous study was <0.2% and the actual distribution in
the current study was 2.0%. For a patient with one predictor, the predicted
probability of septic arthritis from the previous study was 3.0% and the
actual distribution in the current study was 9.5%. For a patient with two
predictors, the predicted probability of septic arthritis from the previous
study was 40.0% and the actual distribution in the current study was 35.0%.
For a patient with three predictors, the predicted probability of septic
arthritis from the previous study was 93.1% and the actual distribution in the
current study was 72.8%. For a patient with four predictors, the predicted
probability of septic arthritis from the previous study was 99.6% and the
actual distribution in the current study was 93.0%.
Receiver Operating Characteristic Curves
Receiver operating characteristic curves for the clinical prediction rule
in the original population and the current population are shown in
Figure 1. The data values for
the original population and the current population are shown in
Table II. The area under the
curve for the original population was 0.96, indicating that this group of four
multivariate predictors demonstrated excellent diagnostic performance in the
identification of septic arthritis. The area under the curve for the current
patient population was 0.86, indicating very good diagnostic
performance31.
Clinical prediction rules are intended to make the art of diagnosis more
objective by allowing the clinician to estimate the probability of a
diagnostic outcome and to classify patients according to the risk of
disease28,29.
Prediction rules originally took the form of clinical aphorisms based on the
empiric experience of senior clinicians; however, more recently, they have
been derived from evidence-based mathematical
analyses28,29.
For the clinician who is faced with the important but often difficult task of
differentiating between septic arthritis and transient synovitis of the hip in
children, the clinical prediction rule that we previously described may be
useful for guiding the diagnostic workup and establishing a timely and
accurate
diagnosis27.
Clinical prediction rules typically demonstrate diminished performance in a
new patient population because they are optimally modeled to the original data
set28,29.
Therefore, the validation of a clinical prediction rule is essential. A
clinical prediction rule can be validated within the original data set by
means of selective sampling, a process known as
boot-strapping28,29.
However, ideally, a clinical prediction rule should be validated by examining
its performance prospectively in a new patient
population28,29.
In the present study, the clinical prediction rule for the differentiation
between septic arthritis and transient synovitis of the hip in children
demonstrated diminished, but nevertheless very good, diagnostic performance in
a new patient population. The same four independent multivariate predictors of
septic arthritis that had been identified in the original population were
found in the current population. The predicted probability of septic arthritis
of the hip from the prediction rule was similar to the actual distributions in
the current patient population. The area under the receiver operating
characteristic curve for the current patient population was 0.86, indicating
that this group of four multivariate predictors had very good diagnostic
performance for identifying septic
arthritis31.
Jung and colleagues performed a study, similar to our original derivation
study, of ninety-seven children with transient synovitis of the hip and
twenty-seven children with septic arthritis of the
hip32. They
identified differences between the two groups with use of univariate analysis
and identified five independent multivariate predictors of septic arthritis: a
temperature of >37°C, an erythrocyte sedimentation rate of >20
mm/hr, a C-reactive protein level of >1.0 mg/dL (>0.10 mg/L), a serum
white blood-cell count of >11,000/mm3 (>11.0 ×
109/L), and a joint-space difference of >2 mm on radiographs.
They developed a different algorithm that was based on the thirty-two
combinations of the five predictors. The area under the receiver operating
characteristic curve for their prediction rule was 0.986.
The limitations of the present study and the original derivation study
include the lack of C-reactive protein data. C-reactive protein has been shown
to have greater benefit than the erythrocyte sedimentation rate for the
diagnosis of septic arthritis in
children32,33.
However, during the study period at our institution, the availability of
C-reactive protein testing evolved from weekly testing to once-daily testing
to routine testing. Thus, the results of pretreatment C-reactive protein
testing were available for only 43% (twenty-two) of the fiftyone patients with
septic arthritis and 39% (forty) of the 103 patients with transient synovitis
in the present study. In order to avoid biases associated with incomplete data
analysis and patient selection, C-reactive protein data were not incorporated
into the analysis of the prediction rule.
As in the previous
study27,
approximately half of the patients who had septic arthritis in the present
study had negative cultures and were considered to have "presumed septic
arthritis." This group with presumed septic arthritis had symptoms
similar to those associated with true septic arthritis, negative cultures, and
a high white blood-cell count in the joint fluid (=50,000
cells/mm3 [=50.0 × 109/L]). As in the previous
study27, the
patients who had true septic arthritis in the present study appeared to be
more sick with fever, chills, an elevated erythrocyte sedimentation rate, and
an altered white blood-cell count differential. In addition, the group with
presumed septic arthritis had a signficantly higher percentage of patients
with a history of recent antibiotic use as well as a significantly higher
percentage of female patients. It is unclear what this presumed septic
arthritis actually represented: partially treated septic arthritis, bacterial
arthritis with organisms that were difficult to grow on culture, viral
arthritis, arthritis resulting from atypical organisms, inflammatory or
rheumatic arthritis, trauma, periarticular osteomyelitis, or an autoimmune
process. Nevertheless, these patients are typically treated identically, with
urgent surgical drainage and
antibiotics30.
In conclusion, the previously developed clinical prediction rule for the
differentiation between septic arthritis and transient synovitis of the hip in
children demonstrated diminished, but nevertheless very good, diagnostic
performance in the current patient population. Further research is needed to
examine its performance in new clinical settings and geographic locations. As
our institution is a tertiary-care children's hospital, differentiating
variables may vary in a community hospital setting. The clinical prediction
rule was developed to differentiate between septic arthritis and transient
synovitis in the typical clinical scenario. Atypical patients, such as those
with immunocompromise, neonatal sepsis, renal failure, postoperative septic
arthritis, and associated proximal femoral osteomyelitis were excluded. Thus,
the prediction rule may not be applicable in these clinical settings. Finally,
it should be emphasized that a clinical prediction rule is not meant to be
used as a rigid guideline or to replace clinical judgment. The goal of this
clinical prediction rule is to aid in the often vexing differentiation between
septic arthritis and transient synovitis of the hip in children by stratifying
patients according to the risk of septic arthritis. Clinical judgment is still
necessary in the further management of these patients. Patients with a minimal
probability of septic arthritis (zero predictors) in the appropriate clinical
setting with close follow-up may be managed with observation only, whereas
patients with a high probability of septic arthritis (four predictors) may be
candidates for aspiration in the operating room instead of the radiology suite
given the greater likelihood that they will require surgical drainage.
Tables showing the results of univariate analysis and the probability of
septic arthritis are available with the electronic versions of this article,
on our web site at
(go to the article citation and click on "Supplementary Material")
and on our quarterly CD-ROM (call our subscription department, at
781-449-9780, to order the CD-ROM).